5 ways to optimize AI costs and reduce wasted AI spend

A 1Password survey of technical workers found that 62% reported gaps in how their company manages AI agents, highlighting the challenge of controlling AI costs as usage-based billing drives up expenses. The article outlines five common sources of wasted AI spend, including overuse of flagship models, unmanaged agents, shadow AI, compromised credentials, and fragmented governance, citing examples such as a Cursor experiment where building a web browser cost $10,565 with a top-tier model versus $1,339 with a mix, and an IBM podcast story of a bill spiking from $180 to $82,000 in two days due to a stolen API key.

The tokenmaxxing era has left companies grappling with an uncomfortable reality. Now that AI vendors have switched to usage-based billing models, businesses are facing sky-high bills, and IT and finance teams are under pressure to rein in spending without slowing down innovation. The logical first step is to locate areas where that spend is going to waste, but even getting visibility into usage can be overwhelming when it’s spread across departments, users, models, vendors, and agents. If you’re trying to track down wasted AI spend and find opportunities to optimize your tokens, it helps to start with some of the primary reasons why AI bills may balloon past your company’s budget. So IT and Finance teams can know where to focus their efforts, here are five of the most common sources of unexpected AI spend. For businesses to optimize spend, they need a way of overseeing and enforcing which models are being used for what tasks. Different AI models can vary wildly both in their abilities and their cost, and many users default to flagship AI models without realizing that there are more affordable options that can accomplish their goals at a fraction of the cost. For instance, in a recent experiment https://cursor.com/blog/agent-swarm-model-economics run by Cursor, building a web browser from scratch cost $10,565 when using a top-tier flagship model, and $1,339 when using a mix of models, even though the end results were comparable in terms of quality. As the Stanford Digital Economy Lab reported https://digitaleconomy.stanford.edu/news/how-are-ai-agents-spending-your-tokens/ , AI agents are “uniquely expensive, consuming 1000x more tokens than code reasoning and code chat.” Meanwhile, Here’s a scenario that’s becoming familiar to many AI developers and builders: An agent is instructed to perform a certain task, but it fails. So it tries again, and fails. With each loop, it gathers more context and uses more tokens than the previous attempt, and nobody thinks to check on it until it’s consumed several engineers worth of tokens literally overnight. IT teams and AI program managers need to ensure agents aren’t allowed to run without oversight from an accountable human, and to build strong harnesses that prevent them from going off the rails. Those unmanaged agents are just one example of “shadow AI,” or AI tools being used without the knowledge or oversight of a company’s IT team. In 1Password’s recent survey of technical workers https://1password.com/blog/survey-ai-agent-adoption-is-outpacing-governance , 62% reported gaps in how their company manages AI agents alone, and Teams and individuals can sign up for these tools outside centralized procurement processes, or may even be using company-provisioned tools for personal projects https://www.infosecurity-magazine.com/opinions/employees-misusing-ai-tools/ , and IT and Finance teams are unaware of them until the bill shows up. In a story told on IBM’s Security Intelligence podcast https://www.ibm.com/think/podcasts/security-intelligence/llmjacking-how-hackers-steak-ai-api-keys , a business had a typical AI bill of $180 a month, but in two days it shot up to $82,000. In this case, the sudden spike had nothing to do with changes to billing models; rather, a bad actor had used a stolen API key to hijack AI compute from the company. Cases like this are just one example of how unsecured and compromised credentials can have unforeseen side effects when it comes to token consumption. As AI compute becomes a more expensive commodity, be on the lookout for more AI-jacking stories, and make sure your company's AI access is always managed and secure. AI governance and spend management is made complicated by fragmented reporting and unclear ownership. 1Password’s recent survey https://1password.com/blog/survey-ai-agent-adoption-is-outpacing-governance found that there’s no consensus among technical employees about who is actually accountable for AI in their organization, a fact that has implications for both security and budgets. Typically, IT teams can monitor managed applications, and Finance teams can see invoices, but both teams have to gather data from multiple dashboards provided by their AI vendors, which may not alert them about potential overages until it’s too late. Without a centralized source of oversight for AI usage and spending, they can accumulate rapidly, unmonitored and unmanaged, until teams receive a bill that nobody planned for. To optimize and reduce their company’s overall AI costs, IT and Finance teams need to collaborate closely to identify and manage unnecessary AI spending. To do so, they’ll need the right tooling. For instance, with 1Password AI Spend Management https://1password.com/solutions/ai-spend-management , teams gain a centralized dashboard to oversee AI use and break down token consumption by vendor, model, team, and user. It also enables controls, such as spending limits and overage alerts, while providing the reporting needed for AI governance and compliance. With essential controls in place to take care of some of the waste, both teams will have the breathing room needed to think more strategically about how their company will govern and optimize AI costs in the years to come. Want practical tactics and tools for AI governance? Read: A practical guide to AI spend management across IT, finance and AI program leaders